logoalt Hacker News

thesz • today at 6:20 PM • 1 reply • view on HN

  > same as interaction nets, no?
Exactly my point. But well researched and has compilers from high level languages (Id and some others).

  > except they are strongly confluent and have incremental sharing
Data flow is an inverted representation of data demand (lazy evaluation, which is strongly confluent) [1] and I do not see why it cannot (be constrained to) have strong confluence.

[1] https://dl.acm.org/doi/pdf/10.1145/93548.93578

Have authors compared interaction nets to the data flow machines? No.

They did not even compared their results to the Transport Triggered Architecture [2] automatic core construction workflow.

[2] https://web.archive.org/web/20051125195128/http://ce.et.tude...

In some cases (FFT), TTA's automatic synthesis was able to provide power-per-operation reduction of three orders of magnitude, when compared to specialized ASIC hardware.


Replies

barthold123 • today at 8:56 PM

I don't know much about data flow but how would lazy evaluation ever be strongly confluent without incremental sharing? Or do you mean not sharing at all, thus not making use of being lazy (in the sense of call-by-name)?